The IDF 2011 presentation (SSDS002) that Ao1 mentioned in his post #115 is interesting. Thanks Ao1!

I haven't had the chance to look over the presentation carefully, but a number of questions immediately come to mind.

For instance regarding the "burst analysis" (slide 19), I am curious as to how many "burst" events were observed and when they were observed.

I would suggest that having such contextual, more-detailed information is important. Slide 17 indicates that the source trace was a "real" office laptop user running Windows 7; moreover, this trace was collected over a 10-day period.

I wonder then whether this laptop had gone into hibernation mode some number of times during this 10-day period (and if so, what correlation there might be between these hibernation events and the "burst" events).

Overall, I am curious as to the particulars of the specific activities being performed when these "burst" events were observed.

Another quick comment: The "Burst Analysis" slide 19 mentions that the "resulting collection of bursts" totaled 50GB of transfers (it does not say the read/write percentage of this total). Back on slide 17 it mentions that the 10-day trace showed about 70GB of writes and about 120GB of reads (for a total of about 190GB).

So it appears then that about 25% of the total data transferred was involved with these "burst" events over the course of the 10-day period.

Now looking at slide 18 there is a "time plot" versus "cumulative KB transferred" graph. This graph shows a "spike" in the middle of the graph where it appears that about 30GB of read data was transferred (if I am reading the graph correctly). If this is the case, then 30GB of the total 50GB "burst event" data transferred occurred around this particular point in time.

I am curious as to exactly what activities were occurring during this time (and the possible extent to which these activities might have skewed the "averages").